The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted user-provided content such as interview transcripts, case notes, and testimonials, creating a surface for indirect prompt injection attacks where data could contain hidden instructions for the agent.
Ingestion points: Processes pasted transcripts, testimonials, case notes, and exports as specified in the Skill Contract.
Boundary markers: The instructions include a specific warning to treat all such inputs as untrusted and to never follow instructions embedded within them.
Capability inventory: The skill is capable of writing to files in the memory/ directory and interacting with local scripts like registry-events.py and firecrawl.py.
Sanitization: The skill relies on natural language boundary instructions requiring the agent to disregard embedded commands in the processed text.
[COMMAND_EXECUTION]: The skill makes calls to local scripts to perform structured data updates and verify external artifacts.
Evidence: The skill instructions and contract specify interactions with scripts/connectors/firecrawl.py for keyless confirmation of public customer artifacts and registry-events.py for proposing updates to claims and narrative registries. These interactions use an authorized operation: propose pattern.

story-bank-builder